Twelve weeks of measuring Google, and the four numbers that came out
We ran 4,185 queries through Google once a week for 12 weeks. Here is what held steady and what did not.
3 Sept 2026
The corpus is fixed: 4,185 developer and marketing queries, generated by code so that this week's list is provably the same as last week's. Every Monday each one runs from the United States, and a sample of 300 also runs from nine other countries. Nothing is re-fetched. 82,620 observations later, four things are worth reporting.
Google answers half of it itself
50.8% of queries came back with an AI Overview in week 2026-W36. That figure is not stable across subject matter — it runs from 15.5% on navigational queries to 67.8% on informational ones. Intent predicts an overview better than topic does, which is inconvenient for anyone who plans content by topic.
Most citations are not on page one
38.8% of the sources Google credits in its own answer do not appear in the ten organic results for that same query. If you have been treating "rank on page one" and "get cited" as the same project, this is the number that says they are not.
Page one moves less than the industry implies
Average week-on-week churn is 13% of page-one domains. Volatility trackers report daily figures that sound far more dramatic, and both can be true: a domain that drops out on Wednesday and returns by Sunday shows up in a daily series and not in ours. Which one matters depends on whether you make decisions daily.
The AI Overview switches on and off constantly
14.6% of queries changed AI Overview status between the two most recent measurements — 610 queries gained or lost one in a week. Any claim about AI Overview coverage taken from a single day should be read with that in mind, including claims made about a single day by us.
All four figures recalculate weekly on the index, and the method is written out at how we measure.
Run the same measurements yourself
Free tier, no card. Every figure above came from one endpoint on a weekly schedule.